On 09/12/16 09:15, Gert Doering wrote:
> Hi,
> On Fri, Dec 09, 2016 at 03:52:32AM +0100, David Sommerseth wrote:
>> - Instead of checking the complete in_addr_t (which lacked proper htonl()),
>>   just do a simple peek at the last byte which contains the first octet
>>   of an IP address or subnet mask.
> Have you *tested* this on a non-intel platform?
> I know that I said on IRC that we might do it that way, but looking
> at stuff like print_in_addr_t() I'm not sure our usage of in_addr_t is
> consistant with "normal usage of it" (namely, always in network byte
> order).

I don't have access to any big-endian machines.  I might still have
access to a virtual POWER8 machine, but that is configured in
little-endian mode - so no difference from the Intel platform.

After I sent this patch, I started to ponder on if there were any
performance reasons why to change the in_addr_t check.  And I've come to
the conclusion I doubt it makes much of a difference.

First of all, an IPv4 address is 32-bits, which is the smallest CPU
variants we do support.  So 0xff000000 or 0xff will just be copied into
a 32-bit register anyway.  Secondly, checking for just 0xff will remove
one AND operation on the CPU and the mov(e) operation will just be
slightly different as it only loads 8 bits from a memory region vs 32
bits.  So we might end up with or without saving just a couple of CPU
cycles.  And when considering that this is code paths which is not
called very often, it starts to smell very much like pointless

I suggest we leave the in_addr_t parsing unchanged, and fix the message
handling to differentiate between --ifconfig and --ifconfig-push.  And
after 2.4.0, we can look at the in_addr_t tackling to see if it behaves
correctly on both big- and little-endian.  If it doesn't behave well
then we'll fix it.


kind regards,

David Sommerseth
OpenVPN Technologies, Inc

Attachment: signature.asc
Description: OpenPGP digital signature

Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today.http://sdm.link/xeonphi
Openvpn-devel mailing list

Reply via email to